Tokens increase on the same prompt from gpt-4o-mini

Just end of August 2024 where the USA is long weekend. I am seeing my software got a lot of error 429 where the TPM exceed or close to the limit. ( I am tier1)

I actually has 1 test prompt which always got 17xx tokens. I try that prompt again on 2nd Sep and found it increase to 35xx tokens on that prompt (in picture, i try 1 prompt 5 times).

However when convert to money on usage, it is still same dollar. However because of that, i got error 429 because of TPM is reach to limit manytimes. So i have to slow down my software many and many times.

Anyone get the same issue with me?
Did the Openai found the same root cause ? Or this is the intend by openai?